tuantc/loupe-1.1

Loupe 1.1

A LoRA adapter for Qwen/Qwen3.5-4B that makes typed decisions about a document in one forward pass and returns probabilities over the options: a System-1 decision engine, fast judgment rather than generated reasoning. Served with the code at github.com/tuantong/loupe-serving (tag loupe-1.1 or later), which reads the answer from the next-token probabilities of the option labels; nothing is generated.

Serving

Merge the adapter into the base weights at load and serve it with LOUPE_PROMPT_STYLE=compact, the template it was trained with. Point LOUPE_CALIBRATION at this repository's calibration.json, which maps the label probabilities to the reported distribution and names the model (loupe-1.1).

Limitations

English only. It decides from what the document says and does not look anything up. Temporal arithmetic over long documents, ranked-rule tradeoffs and long policies remain its weakest families.

Licence

Apache-2.0, like the base model (Qwen/Qwen3.5-4B) and the serving code. The training data was written by DeepSeek V4.1 Flash, with targets from Qwen3.8-27B (Apache-2.0) and DeepSeek V4.1 Flash.